Prenatal Risk Factors for Cutaneous Malignant Melanoma: Follow-up of 2,594,783 Danes Born from 1950 to 2002

The incidence of cutaneous malignant melanoma has increased over past decades in most young adult populations of European origin (1-4), although signs of leveling off have been detected in Northern Europe (4). The etiology is hypothesized to be related in part to prenatal exposure to hormones, particularly estrogen (5, 6). This study focuses on the occurrence of cutaneous malignant melanoma among young people in Denmark, with increasing incidence over calendar time up to 2002, particularly in females. It remains unclear whether prenatal exposure to hormones plays an important etiologic role in this trend (3). Data on the extent of estrogen's effect on cutaneous malignant melanoma incidence are divergent (2, 3, 7-10), and most recent review papers do not even mention estrogen (1, 11).

It has been shown that transplacental exposure to diethylstilbestrol increases the risk of dermal melanomas in hamster progeny (12), with estrogen receptors expressed in their skin (13). Whether estrogen exposure in adult life or in utero modifies the risk of cutaneous malignant melanoma in humans remains unsolved. Estrogen receptors have been found in conjunctival cutaneous malignant melanoma, and estrogen exposure in adult life has been associated in some studies with cutaneous malignant melanoma risk, although a mechanism for cancer induction has not been identified (14).

Postponement of reproduction could explain some of the increase in cutaneous malignant melanoma incidence. Higher age at childbearing, specifically the first birth, may increase the mother's risk of cutaneous malignant melanoma (15-17), and could also affect the fetus. Neale and colleagues found that offspring of mothers in the 40-49-year age group had an estimated relative risk of cutaneous malignant melanoma of 1.29 [95% confidence interval (95% CI), 1.03-1.63] compared with offspring of mothers <20 years (15). Because twin pregnancies are characterized by a high level of estrogen, one might expect a higher risk of cutaneous malignant melanoma in twins than in singletons if estrogen is a risk factor. Existing data, however, do not support this hypothesis (8).

Cutaneous malignant melanoma has two sets of relatively well-established risk factors. The first set involves sun exposure and skin characteristics such as pigmentation and number of moles (2, 7, 9, 10). Very frequent sunbathing, common among adolescent girls, is thus a possible explanation for the higher cutaneous malignant melanoma risk observed among young women of reproductive age compared with young men (3). The second set of risk factors is genetic: CDKN2A and CDK4 mutations are associated with cutaneous malignant melanoma occurrence (3, 18).

The use of data from Danish population registries allows the long-term follow-up required for prospectively examining the association between pregnancy-related conditions and the risk of cutaneous malignant melanoma. The registries also allow linkage of parent and offspring data, permitting control for a family history of cutaneous malignant melanoma (7, 18). We obtained data on cancer incidence and birth characteristics at the individual level through linkage of different population registries available in Denmark.

We examined the hypothesis that cutaneous malignant melanoma is more frequent in persons who had prenatal indicators of high levels of estrogen exposure. This applies to twins, children with high birth weight (19), preterm birth, some congenital malformations, and maternal hyperemesis during pregnancy (5, 6). We also considered cutaneous malignant melanoma occurrence in offspring of mothers who developed preeclampsia during pregnancy, as this condition is an indicator of a low uterine estrogen level (5, 6). To our knowledge, evidence for the magnitude of the associations between those estrogen indicators and cutaneous malignant melanoma risk in offspring is very sparse. Family characteristics such as high sib order and large family size were considered possible protective factors (15, 20). The study by Hemminki and Mutanen suggests decreasing risks for both high birth order and large family size (current size of sibship). After mutual adjustment, large family size exerted a larger effect than mother's parity (approximate 30% reduction for family sizes of 5 to 17 children versus 1 child; ref. 20). Besides exploring the hypothesized associations between prenatal hormonal risk factors and later cutaneous malignant melanoma risk, we also addressed the occurrence of cutaneous malignant melanoma and the mutual effects of maternal characteristics.

We conducted this nationwide historical follow-up study using prenatal data from hospital records and birth records. Incident cutaneous malignant melanoma cases were identified from the Danish Cancer Registry [International Classification of Diseases (ICD-7) code: 190]. In women, we considered the following cancers as possibly hormone-related: breast cancer (ICD-7 code: 170), cancer of the corpus uteri (ICD-7 codes: 172, 173, and 174), and ovarian cancer (ICD-7 code: 175). In men, we considered prostate cancer (ICD-7 code: 177) and testicular cancer (ICD-7 code: 178). These encompass incident primary carcinomas, including certain special cases, and metastases from the cancer in question. For testicular cancer we also included embryonal carcinoma, seminoma, and testis teratoma.

Denmark's Civil Registration System (CRS) assigns to all Danish citizens a health insurance card with a unique personal identification number (21). This number is recorded at all hospital admissions. It also is used to register all births, deaths, migrations to and from foreign countries, and domestic address changes. The Danish National Board of Health operates several health-related nationwide health registries: the Medical Birth Registry (since 1973), the Registry of Congenital Malformations (1983-1996), the Cancer Registry (since 1943), and the National Hospital Discharge Registry of Patients (since 1977). The Discharge Registry includes all nonpsychiatric hospital discharges, and, since 1995, visits to outpatient clinics and emergency rooms. The personal registration number allows individual-level linkage among the various Danish administrative registries. This study was approved by the Danish Data Protection Agency, which requires adherence to a number of data storage rules to avoid disclosure of personal data.

Using the CRS, we identified all women born after 1935 who were alive on April 2, 1968. We focused on women born after 1935 to maximize identification of linkage among sibs, and their common mother in the CRS, and thus to control for familial aggregation of cutaneous malignant melanoma. Because the CRS was established on April 2, 1968, the Registry included only citizens alive on that date at its initiation. After identifying this cohort, we obtained information on our study population, composed of all children born to the women from 1950 to 2002, and alive on April 2, 1968. Follow up continued from birth until cutaneous malignant melanoma occurrence, or death, emigration, or the end of the study period on December 31, 2002. The study outcome comprised all incident cutaneous malignant melanoma cases among offspring born after 1950, identified from the Cancer Registry. The outcome measure was the incidence rate ratio (IRR) of cutaneous malignant melanoma, estimated using Poisson regression in STATA 9 (22). Most of the covariates (i.e., exposures and potential confounders) were treated as time-dependent variables. Other exposures and confounders known at birth were fixed throughout the follow-up period. Information on date of birth, sex, multiple births (twins, triplets, and higher order births were pooled), sib order, family size, and mother's age, together with family links to sibs (and mothers), were available from the CRS for the 1950-2002 period. Data on preeclampsia and hyperemesis in the mothers were available from 1977 to 2002, whereas information on preterm birth, birth weight, and congenital malformations (with date of detection) was available from 1973 to 2002. Because of the dynamic-cohort design and expected age-period effects, we subdivided the time-at-risk by age (in age groups: 0-14, 15-19, 20-24, 25-29, 30-34, 35-39, and ≥40 y) and calendar time (based on four periods: 1950-1974, 1975-1984, 1985-1994, and 1995-2002). Analyses were restricted by the time periods for which data were available. In the multiple regression analysis, we included all variables, making it possible to interpret their regression coefficients as estimates of corresponding main effects mutually adjusted for the other variables. Records with missing data on birth weight and weeks of gestation were included in the reference category because alternative ways of handling the missing information did not affect the results. We obtained information from Statistics Denmark on maternal gross income (in four groups separated by quintiles), marital status (single, married/cohabiting), employment status (self-employed, salaried employed, pensioner), educational level (university degree, short/medium formal education, basic vocational education, other), and degree of urbanization (capital, provincial town with >10,000 inhabitants, rural area). These variables were available for the 1980-2002 period.

Based on the hypothesis concerning a potential genetic component, we expected some clustering of melanoma cases in families (sibships). To address this issue, we recalculated the SE of estimates using the Huber-White method for clustered data (23).

Among 2,594,783 persons born during the 1950 to 2002 period, we identified 1,674 incident cases of cutaneous malignant melanoma during 56.3 million person-years at risk. Among females, 1,118 cutaneous malignant melanoma cases were observed during 27.3 million person-years at risk, whereas among males 556 events were observed during 29.0 million person-years at risk. Table 1 provides the distribution of birth characteristics among offspring.

Initially, we estimated the basic multivariate model shown in Table 2. This model included the variables for family size, sib order, mother's age, mother's age at first birth, multiple birth, history of cutaneous malignant melanoma in full sibs or mother, sex, age, and calendar period. We then excluded the variables one by one if the P values from likelihood ratio tests and the relative risk estimates of a given variable suggested lack of association with the risk of cutaneous malignant melanoma at the 5% level. This led to the elimination of mother's age at birth (P = 0.57), multiple birth (P = 0.18), and the sib order variable (P = 0.15). The reduced model produced estimates similar to those of the basic model. However, we noted that the effects of family size and mother's age at first birth were slightly enhanced in the reduced model. The full multivariate model in Table 2 was based on published findings, and was therefore not data-driven like the selective reduced model. Future cutaneous malignant melanoma investigators thus could use the evidence from our study and choose to apply the reduced model (24). The only pairwise interaction statistically significant at the 5% level was that between age and sex (P = 0.04), but analyses with or without this interaction term yielded similar results. The risk was modified by sex through an excess aggregation of cases among women ages 25 to 29 years, and by an excess risk in male offspring from the age of 35 years on.

We found that belonging to a large family conferred reduced risk of cutaneous malignant melanoma, at least in families with five children or more, with IRR at 1.46 (95% CI, 1.01-2.10) for families with 1 child, 1.62 (95% CI, 1.18-2.25) for 2 children, 1.60 (95% CI, 1.16-2.20) for 3 children, and 1.50 (95% CI, 1.07-2.09) for 4 children, each compared with families with ≥5 children. Controlling for sib order did not affect the estimates as long as family size was included in the model. Family history of cutaneous malignant melanoma in full sibs was associated with a 4-fold increase in risk, whereas family history of cutaneous malignant melanoma in the mother conferred a 3-fold increase in risk. After adjustment for family history of cutaneous malignant melanoma, the family history of hormonal cancers (excluding cutaneous malignant melanoma) in both full sibs and the mother was not associated with a greater cutaneous malignant melanoma risk. The risk of cutaneous malignant melanoma in a given child increased most with the mother's age at first birth and less with the mother's age at the birth of that child.

We found no large increase of cutaneous malignant melanoma risk in twins compared with singletons (Table 2), although a larger study would be required to detect a moderately increased risk. There was no modification of the effect of twinning by sex (P = 0.32) in the basic multivariate model. Table 3 provides results of the detailed analyses of twins with a co-twin of different sex versus same sex. There were no statistically significant differences in the cutaneous malignant melanoma risk (at the 5% level) between singletons and co-twins of any type. The relative risk of cutaneous malignant melanoma in male twins was similar to the risk for male singletons, both when the co-twin was of the same sex (IRR, 1.03; 95% CI, 0.46-2.33) and when the co-twin was of the opposite sex (IRR, 0.78; 95% CI, 0.19-3.13). In contrast, female twins with male co-twins tended to have an increased cutaneous malignant melanoma risk (IRR, 1.80; 95% CI, 0.96-3.39) compared with female singletons. Female twins with co-twins of the same sex had no increased risk of cutaneous malignant melanoma compared with female singletons (IRR, 1.24; 95% CI, 0.74-2.08).

Table 3 shows the results of our examination of the effects of twinning, preterm birth, high birth weight, preeclampsia, hyperemesis, congenital malformations, and histories of hormone-related cancer in full sibs and mothers—one variable at a time in the basic model in Table 2. Due to paucity of data, the precision of estimates is low.

Among the 2,594,783 children born from 1950 to 2002, we identified 1,217,527 family clusters of sibs with the same mother. To assess the extent of exaggeration caused by failing to model correlation between family members, we refit the basic multivariate model both controlling for and not controlling for correlation using a 30% random sample of the family clusters. The effect on SE and CI of the estimated hazard ratios was negligible, and precision was in some cases better in the analysis that did not control for the correlation.

In the period from 1980 to 2002, we had information on the socioeconomic conditions of the mothers at the birth of each offspring. However, the number of cutaneous malignant melanoma cases was limited and we were not able to control for that confounder. In a supplementary analysis, we investigated the associations between sib order and available socioeconomic variables in the 1980-2002 period. We compared socioeconomic status among mothers of children of sib order 5 or more to that among all mothers, and found in the former population more pensioners and economically inactive women (excluding unemployed persons with benefits and coworking wives; 48.2% versus 20.6%), more single mothers (14.2% versus 7.8% on average) with no qualifying education (67.4% versus 58.9%), living in a rural area (51.3% versus 34.1%), and belonging to the lowest quintile of gross income (48.6% versus 19.9%).

Our data provide evidence that mother's age at first birth, family history of cutaneous malignant melanoma in full sibs and mothers, child's sex and age, and calendar period of birth are all associated with increased cutaneous malignant melanoma risk, whereas large family size is associated with decreased risk. Markers of prenatal hormone levels (preterm birth, high birth weight, congenital malformations, preeclampsia, and hyperemesis) did not seem to be strong risk factors for cutaneous malignant melanoma. Changes in family size and older age at the beginning of reproduction provide one possible explanation for increasing cutaneous malignant melanoma incidence over time.

To the best of our knowledge, this study is the first to focus on family history of possible hormone-related cancers and cutaneous malignant melanoma risk. Our analyses controlling for family characteristics, sex, age, and calendar period showed that family history of cutaneous malignant melanoma increased the risk of cutaneous malignant melanoma by a factor of 3 to 4. Additional adjustment for family history of other possible hormone-related cancers in full sibs or in the mother did not affect the relative risk of cutaneous malignant melanoma in offspring. Analyses of family history of hormone-related cancers alone, omitting family history of cutaneous malignant melanoma, yielded similar estimates as those reported (data not shown). A possible increase in estrogen levels does not explain the increasing cutaneous malignant melanoma incidence (25).

Our data corroborate the hypothesis of an inverse association between family size and the risk of cutaneous malignant melanoma (20, 26, 27); this might be explained by fewer opportunities in large families to travel to holiday locations with more sun exposure than Denmark. Compared with children of low sib order, children of sib order 4 or higher were more likely to be born in the least urbanized areas, and slightly more likely to have single mothers with low income and little education. Children of high sib order implicitly belong to large sibships with the same mother, but our study suggests that large family size is a more important risk factor for cutaneous malignant melanoma than high sib order. We cannot, however, rule out the possibility of high sib order being associated with reduced risk of cutaneous malignant melanoma as indicated by the marginally statistically significant risk ratio of sib orders of 5 or more compared with a sib order of 1, in the basic multivariate model including adjustment for family size.

The effect of family size on the risk of cutaneous malignant melanoma could be mediated through sun exposure. Postponed reproduction might be related to certain lifestyles and subsequently be mediated through social differences in access to sun exposure. Because large families and other less well-off families cannot afford frequent sun tourism, they are protected against high sun exposure, particularly against the exposure of sensitive skin areas such as the trunk. Incidence rates for melanoma on the trunk have increased markedly since the 1960s (7). The review by English et al. concluded that cutaneous malignant melanoma occurrence is more strongly associated with nonoccupational than with occupational sun exposure, the latter primarily affecting the head, face, and neck (9, 28). To our knowledge, evidence on the joint influence of family socioeconomic status and sun-seeking behavior on cutaneous malignant melanoma risk is limited. In the period for which data on socioeconomic factors were available (1980-2002), we found an inverse relation between sib order and maternal socioeconomic status, a pattern also found in other studies (10, 20, 26). We examined an earlier Danish survey that collected data on the number of children <15 years in the household, household income (in quartiles), and self-reported use of tanning beds and frequency of getting an outdoor sun tan (29). The survey, based on interviews of 1,000 Danes ages 15 to 90 years, did not support any associations between sun-seeking behavior and family size (P = 0.49 for use of sun beds; P = 0.53 for frequent tanning). Neither was the use of tanning beds associated with household income (P = 0.12), with 25% (95% CI, 22-29%) of nonusers versus 22% (95% CI, 17-27%) of users in the 1st (lowest) income quartile, 19% (95% CI, 16-22%) of nonusers versus 17% (95% CI, 13-21%) of users in the 2nd quartile, 20% (95% CI, 17-23%) of nonusers versus 27% (95% CI, 22-32%) of users in the 3rd quartile, 28% (95% CI, 24-31%) of the nonusers versus 27% (95% CI, 22-32%) of users in the 4th quartile, and 9% (95% CI, 7-11%) of nonusers versus 8% (95% CI, 5-11%) of users in the group of participants that did not report on this question. Frequent sun tanning was more common in the lower quartiles of household income (P = 0.007), with 35% (95% CI, 28-43%) of nontanners versus 22% (95% CI, 19-25%) of tanners in the 1st (lowest) income quartile, 16% (95% CI, 11-23%) of nontanners versus 18% (95% CI, 16-21%) of tanners in the 2nd quartile, 19% (95% CI, 14-26%) of nontanners versus 27% (95% CI, 22-32%) of tanners in 3rd quartile, 22% (95% CI, 16-29%) of nontanners versus 28% (95% CI, 25-32%) of tanners in the 4th income quartile, and 7% (95% CI, 4-12%) of nontanners versus 9% (95% CI, 7-11%) of tanners in the group of participants that did not report on this question. Subsequently, we investigated the association between income quartile and the number of children under 15 years, if any children were in the household. Contrary to our expectation, we found no tendency for low-income households to contain more children. It remains an open question whether differences in other lifestyle patterns or selection problems could make this finding fit with the tendency of offspring of high sib order to have less well-off mothers.

Inclusion of the mother's age at first birth eliminated the effect of mother's age at the time of a given child's birth, although this may be an artifact of the strong correlation between the two variables. An increased rate of mutagenesis with age may explain the possible effect of the mother's age, as suggested for some other cancers (6). However, the fact that age at first birth is the most significant risk factor for cutaneous malignant melanoma raises the question of whether later age of reproduction explains some of the increased cutaneous malignant melanoma risk in women age <45 years. Maternal subfecundity reflected through later age of reproduction could also be associated with high cutaneous malignant melanoma risk in offspring. We did not find a significant interaction between mother's age at first birth and calendar period, suggesting that the effects of late age of first birth remained stable over calendar time. However, more data are needed to clarify this issue. We did find a clear trend for increasing cutaneous malignant melanoma risk with mother's age at first birth (P = 0.0007). In a Swedish registry-based study of the risk of cutaneous malignant melanoma in mothers, similar results for the effects of age at first birth and parity (family size) were reported (15). However, one recent case-control study reported increasing risk of cutaneous malignant melanoma in mothers by parity (30).

In our study, the IRR of cutaneous malignant melanoma of 1.21 (95% CI, 0.86-1.72), comparing twins with singletons, was in the opposite direction from the estimate of 0.67 (95% CI, 0.42-1.06) reported by Neale and colleagues (ref. 8; P = 0.02, testing the hypothesis of equality of rate ratios in the two studies). This discrepancy may be due to chance or could be caused by differences in cutaneous malignant melanoma definitions or age limits. We used a narrow definition of the outcome, limiting it to primary and metastatic cutaneous malignant melanoma. Lymphoma, sarcoma, and carcinoma in situ were excluded to focus on the possible hormonal etiology.

One limitation in our study is the lack of control for paternal characteristics. We cannot rule out the possibility that cutaneous malignant melanoma in offspring is affected by paternal age or paternal history of cutaneous malignant melanoma, as suggested by previous studies (31, 32).

Socioeconomic conditions are mentioned in the literature as a confounder in studies of cutaneous malignant melanoma occurrence, especially in discussions of the decreased risk of cutaneous malignant melanoma in children from large families (20). In the subset of data with available information on socioeconomic factors (1980-2002), the number of cutaneous malignant melanoma cases was too sparse to pursue this analysis. Our examination of associations between high sib order and maternal socioeconomic status clearly showed that being born into a large sibship is associated with lower maternal social class. It is noteworthy, however, that a recent large study by Lawlor and colleagues suggested that the risk of cutaneous malignant melanoma is similar among children of parents of manual and nonmanual occupations (33), leaving open the question of whether low social class confounds the association between family characteristics and the risk of cutaneous malignant melanoma.

A family history of cutaneous malignant melanoma could conceivably induce other family members to use sun screen or other protection against the sun. Such behavior could reduce the risk of cutaneous malignant melanoma in family members, and thus result in underestimation of the true effect of family history on cutaneous malignant melanoma risk. Another important issue that calls for further research is the possible interaction between genetic factors and sun exposure (7, 34, 35).

Our lack of control for sun exposure could have biased the estimated effect of large family size. Other potential risk factors might have contributed unexplained variation due to sun exposure, and thus have caused spurious effects, as in the example of family size. As well, some estimated associations might be artificial time trends of sun exposure level that has increased in succeeding cohorts. In Denmark, along with more frequent sun bathing among young women, parity has decreased and age at mother's first birth has increased. When these variables are introduced into the model, they could reflect time trends parallel to that of sun exposure.

The information on outcome and exposures in this large nationwide registry-based data set can be considered reasonably accurate. However, nondifferential misclassification of outcomes is expected for data obtained from the Cancer Registry, as well as for exposure data obtained from the Medical Birth Registry and the Hospital Discharge Registry of Patients. Such misclassification may have attenuated our effect estimates.

Our data indicate that familial occurrence of possible hormone-related cancers or exposure to high or low levels of estrogen in utero do not increase the risk of cutaneous malignant melanoma. Either hormones play a minor role in the etiology of cutaneous malignant melanoma, or cutaneous malignant melanoma is influenced by other hormones along pathways other than those involved in breast, ovarian, corpus uteri, testicular, and prostate cancer. Our data suggest that changes in fertility may provide some important clues in understanding this cancer. It is important to elicit the mechanisms by which the mother's increased age at first birth affects the risk of cutaneous malignant melanoma in offspring. To answer this question, study cohorts are needed with available data on fecundity, such as data on insemination use.

It is unclear why children from large families have a lower risk of cutaneous malignant melanoma. This estimated effect may be a bias induced by lack of adjustment for sunburns and other indicators of sun sensitivity. Disentangling these effects would illuminate our understanding of cutaneous malignant melanoma.

In conclusion, this study shows that family characteristics and changes in the reproductive pattern are associated with cutaneous malignant melanoma risk and may in part explain the time trend of increasing cutaneous malignant melanoma in some countries.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Søren Friis at the Danish Cancer Association for his assistance in defining melanoma and hormone-related cancers.